Monday, 17 October 2016

Irish EPA grossly overestimate fatal cancer risk from radiation.


The Irish EPA grossly overestimate fatal cancer risk caused by radiation. By up to 6000-fold for age cohort 20-24. For example, if you are aged between 20 and 24, the Irish EPA overestimate your risk of dying of a fatal cancer due to radiation by 6000 times too much. The real risk is about 1 in a million. Irish EPA estimate it at 1 in 178.

In their fact sheet on radiation[1], The Irish Environment Protection Agency say:

  1. "we can estimate that a dose of 10 µSv may increase the lifetime risk of fatal cancer by about one in 2,000,000"
  2. Their estimate is based on real world risk assessments using:
    • Hiroshima and Nagasaki bomb survivor data
    • Patients exposed to external radiation for the treatment or diagnosis of certain diseases
    • Marshall Islanders exposed to severe fallout from atmospheric nuclear weapons tests
    • Miners exposed to radon and its decay products
    • Residents exposed to radon in the home
    • Workers exposed to radium-226 in luminous paint
    • Patients exposed to radium-224 for bone disease

This EPA leaflet looks convincing. What could be more believable than a risk assessment derived from real world cancer mortalities? Produced by a government agency called the Environment Protection Agency. You would have to take that seriously, or should you?

For a risk assessment to make any sense it should give some estimate of risk within reasonable error bounds. A risk assessment which is out by an order of magnitude (ten times too high or too low) is of little help. Surely an Environmental Protection Agency should be able to get their guesses right within an order of magnitude?

Reasonable explanation proposed for the Irish EPA statement

Joris van Dorp: If 10 muSv/yr gives 1 in 2 million chance of cancer, then assuming 70 years average life, dose is 2.7mSv × 70 yrs ~ 200.000 muSv, means chance of death is 20 thousand in 2 million ~ 1%.

1% is 1/40 of normal cancer incidence of 40%.

Joris' explanation looks good to me. But there's still a problem with this EPA handout. Taken literally it implies a far greater risk than the evidence shows. Irish EPA can should say each '10 µSv per year' if that's what they mean.

My model

I took what I know of regulations and cancer to make a model, which I compared to real world cancer data. Because I live in UK and have real world UK cancer mortality data, it was easiest to model this for the UK. However I can assure you, there's nothing special about UK with regard to cancer risk. If anything UK has a bad rep for stopping deaths from cancer. This is my final model compared to the real world.

Table 1: UK cancer mortalities. Real world compared to projected data
per 100,000
real numbersderivedprojections
Age RangeMale DeathsFemale DeathsMale RatesFemale RatesAverage (MF) ratesAverage (MF) rates - radiationinceptiondeathsOverestimate
0 to 0447392.322.150.12033.667 ×
05 to 0947402. ×
10 to 1438332.11.920.1878117.52349 ×
15 to 1964473.22.52.850.11,215299.54204 ×
20 to 24877543.53.750.11,553562.35998 ×
25 to 2913115166.96.450.21,890870.35397 ×
30 to 342102949.813.611.70.32,2281,197.74095 ×
35 to 393194601622.919.450.52,5651,531.83150 ×
40 to 4467898430.643.4370.92,9031,867.72019 ×
45 to 491428176961.974.668.251.73,2402,204.71292 ×
50 to 5425572853118.9129.9124.43.13,5782,542.2817 ×
55 to 5943604076234.7214.1224.45.63,9152,879.7513 ×
60 to 6473586070422.1334.3378.29.54,2533,217.2340 ×
65 to 69110898696658.1488.2573.1514.34,5903,554.7248 ×
70 to 741259997291043.6724.8884.222.14,9283,892.2176 ×
75 to 7914330112921501.6991.61246.631.25,2654,229.7136 ×
80 to 8414127121942169.51355.91762.744.15,6034,567.2104 ×
85 to 89104551025330311732.42381.759.55,9404,904.782 ×
All Ages8566776644271.6235.2253.46.3


Table 1 shows cancers grouped by age cohort give by UK Cancer Research[2] (yellow). Two columns are derived from real world data (green). The first simply averages male and female fatal cancer rates per 100k. The second derived column divides this by 40 on the assumption that 1 in 40 cancers are caused by radiation and that there's nothing exceptional about radiation cancers compared to others. Next we move to derived data (salmon). The inception column shows the number of cancers which are expected to eventually result in fatalities. The 'deaths' column applies a calculation based on the chart below which accounts for the time taken for a cancer to kill. This shows the number of fatalities expected in that age cohort. The final column (Overestimate) shows the ratio of column 9 (deaths) to column 7 (Average (MF) rates - radiation). Rates in the table are all give as per 100,000

Chart 1 : Cancer induction time distribution[3]


This part of the blog gives a detailed derivation of the previous model.

The Irish EPA say: "we can estimate that a dose of 10 µSv may increase the lifetime risk of fatal cancer by about one in 2,000,000". Nearly all regulatory agencies in the world assume a linear no-threshold effect due to radiation. [ I think the French[4] alone are different ]. So 10 µSv = 0.01 mSv. 1 in 2 million is 0.05 in 100,000. We will compare fatal cancers per 100,000 of the population. If we expect 0.01mSv to give a rate of 0.05, we can project 2.7mSv to cause a rate of 13.5. By simply scaling.

Table 2:
dose (mSv)Projected fatalities per 100k

2.7mSv is the average annual radiation exposure found in UK. Our real world cancer data comes from UK Cancer Research is also refers to the UK. I use this Irish EPA risk assessment to project the inception time of fatal cancers. I get the table below. It increases on a linear scale. Each extra year of life, adds an extra risk of contracting a fatal cancer from background radiation. A risk corresponding to an extra 13.5 per 100,000, per year.

Table 3: Project of fatal cancer inception times.
Year of lifeFatal cancer inception / 100k


Table 3 Notes
  • the table continues, ending at 89
  • Each 5 year period will be summed to make 5-year cohorts. The first cohort has 5 years labelled 0 - 4.

How very simple. I think this is one of the reasons why Linear No-threshold, LNT, is beloved of regulators. It is so very easy to math. Ref [4], has a good explanation of LNT. We can not just go from cancer inception to predict fatalities. There is a delay between inception and mortality which can be quite long (see Chart 1 above). At this point I simplified. I have real world data in 5 year cohorts, and inception-to-mortality data in 5 year cohorts. I grouped my inception data into 5 year cohorts too. By summing the 1-year cohort projections. Because I compare this with fatalities grouped into 5 year cohorts, I sum each 5 year band. The inception time for a fatal cancer differs from the time of death, according to a distribution shown in Chart 1: This chart was used to make a table. (Table 4). The distribution for the last 2 age ranges was smoothed. ( So 1 + 1, rather than 0 + 2). 281 is the estimated sample size. The estimated fatal cancer inceptions for each 5 year cohort were now multiplied to get the estimated time of death. These are the numbers seen under the age ranges (horizontally) These numbers were summed for each age range to arrive at final estimates of actual deaths per cohort.

Table 4: Cancer induction time distribution
total0 to 0405 to 0910 to 1415 to 1920 to 2425 to 2930 to 3435 to 3940 to 4445 to 49
Table 5: How induction time was added
Per 5 yearscancer inceptiondeaths
0 to 0420343.603
05 to 09540289.60918.737
10 to 1487811715.61449.96451.886
15 to 191,21530021.61981.192138.36358.372
20 to 241,55356227.625112.420224.840155.65841.797
25 to 291,89087033.630143.648311.317252.945111.45917.295
30 to 342,2281,19839.635174.875397.794350.231181.12146.1217.927
35 to 392,5651,53245.641206.103484.270447.518250.78374.94721.1391.441
40 to 442,9031,86851.646237.331570.747544.804320.445103.77234.3513.8430.721
45 to 493,2402,20557.651268.559657.224642.091390.107132.59847.5626.2461.9220.721
50 to 543,5782,54263.657299.786743.701739.377459.769161.42360.7748.6483.1231.922
55 to 593,9152,88069.662331.014830.178836.664529.431190.24973.98611.0504.3243.123
60 to 644,2533,21775.667362.242916.655933.950599.093219.07587.19813.4525.5254.324
65 to 694,5903,55581.673393.4701003.1321031.237668.754247.900100.40915.8546.7265.525
70 to 744,9283,89287.678424.6981089.6091128.523738.416276.726113.62118.2567.9276.726
75 to 795,2654,23093.683455.9251176.0851225.810808.078305.552126.83320.6589.1287.927
80 to 845,6034,56799.689487.1531262.5621323.096877.740334.377140.04423.06010.3299.128
85 to 895,9404,905105.694518.3811349.0391420.383947.402363.203153.25625.46311.53010.329

Model derivation explained

  • The cancer inception column in table 5 is derived by summing each consecutive 5 years from Table 3.
    E.g. 203 = 13.5 + 27 + 40.5 + 54 + 67.5
  • This inception data (e.g. 203) is distributed according to the frequency shown in Chart 1 (same as Table 4).
    E.g. 203 = 3.603 + 18.737 + 51.886 + 58.372 + 41.797 + 17.295 + 7.927 + 1.441 + 0.721 + 0.721
  • The 3rd column: deaths, is got by summing the delayed mortalities to the right of it
  • Data after the 85 to 89 column is ignored. It's just displayed to show the model.
  • Columns 2 and 3 of Table 5 go to make columns 8 and 9 of Table 1


  1. The document claiming 10 µSv exposure implies 1 in 2 million mortalies was authored by the Radiological Protection Institute of Ireland, RPII, which was established in 1992 and merged with the EPA in 2014. The EPA still use RPII fact sheets when dealing with radiation.
  2. UK Cancer Research. Download the spreadsheet data.
  3. Source: Chapter 2 - "Epidemiology Kept Simple: An Introduction to Traditional and Modern Epidemiology", by B. Burt Gerstman, Wiley, 2013, page 36
  4. Dose-effect relationship and estimation of the carcinogenic effects of low-doses of ionizing radiation, by Maurice Tubiana, André Aurengo, 2005

No comments:

Post a Comment